DIY-SMU Project

[Kleinstein]

The low current ranges are some of the more sensitive cases when it comes to mains hum and shielding is not allways perfect. 60-70 dB hum suppression is already not so bad. The effort for an external clock is still moderate and could be worth a try. There is anyway a limit in how stable the mains frequency is. A more stable clock could also help with noise and linearity. Just a crystal may react to disturband and could even make things worse, as it is relatively easy to disturb a bare crystal.

[Roehrenonkel]

Hi Dave,

Thanks, Roehrenonkel, you are brave With internal clock the AD7190 50/60 Hz rejection is pretty good at 60-70dB. I did some quick tests a while back to operate the ADC at non-power line frequencies and the noise was still good.  Generally an SMU is not used in a heavy industrial environment or with long cables to low-voltage sensors, no? More of a quiet lab instrument.  Do I underestimate where SMUs are used?
Thanks
Dave

Please keep in mind that we are talking about 144,72 dB (24(Bits) * 6.03 dB) here, so every decibel is precious - don't let them escape. ;-))

Have you tried the sequenzer-function of the AD7190?
In what mode does the ADC run? Chop disabled, Gain=1?

SMU in quiet labs: objection your honor.
Maybe true if used only "stand-alone" for resistance-measurements.
But imagine the Low-terminal (agnd) riding on an ac-voltage or a dc-sweep 0..300V.
My tube-test-system (mainly Triodes) is fairly simple, but where is the quiet spot? ;-))

Thank you, keep up the good work.

[Roehrenonkel]

Hello Kleinstein,

The effort for an external clock is still moderate and could be worth a try. There is anyway a limit in how stable the mains frequency is. A more stable clock could also help with noise and linearity. Just a crystal may react to disturband and could even make things worse, as it is relatively easy to disturb a bare crystal.

The effort is indeed small: 5*7mm and 30mA on the +5V-rail for the smd-oszilator.


Mains-f-stability is +-200 mHz in the EU else the grid shuts down.
See this nice website:
https://www.netzfrequenzmessung.de/
https://www.netzfrequenzmessung.de/verlauf.htm
For ze englisch:
https://www.mainsfrequency.com/
https://www.mainsfrequency.com/verlauf_en.htm
Have made some local test (Tektronix DM5120, DC5009) also:
(yellow=diff. to 230Vrms left scale, blue=diff. to 50Hz right scale)
 Netz-U-f.pdf (128.24 kB - downloaded 83 times.)
To my surprise still good voltage. :-)
Just the wave-form looks more like a trapezoid than a sine.

[SebastianH]

Hi Dave,

a few questions for the pros 

1. you already discussed using an INA for the current measurement. But I also thought about the high impedance feedback to the capacitive JFET input of the OPA*145 and potential ringing. My quick ltspice simulation definitely shows a substantial gain peak and I'd want to avoid those thing in the feedback. The best thing to do would be just using a INA configuration and 10k resistors, which would significantly reduce (not completely avoid..) the peaking (and some noise), I would guess?! One could use a 2145 and use the second opamp as a voltage follower for CLM (post U11.2/3)... Otherwise: Putting 10p accross the R47 and R56 would fix this dynamic behaviour in my sim. Did you observe something like this in reality?

2. In the 1mA range the on-resistance of the DG444 would be roughly 1%. Although this itself is not concerning (if we can ignore things like R_DS(on) vs. V_D, which we might), but the tempco of this resistance is also relatively large, ~15 Ohm difference between 0°C and 85° for the Vishay DG444.

3. Please correct me if I'm wrong: In its current form the provision for the 1A range seems to be not really usable. I don't know of Photo-MOS relays capable of switching 1A while having a specified max. leakage current <<1µA, so it might make sense to use U17.4 for the sensing of the 1 mA range. In case a 1A range shall be implemented, we would more likely use a mechanical relay like the EC2/EE2 series and would have two poles that we could use for both the "sense" and the shunt switching, hopefully without adding too much leakage. Right?

4. 100mA current measurement: Would an array of - say - 8 film resistors, 1206 10ppm 0.4W, be an alternative? Due to the physical size of the array it could be easier to extract the heat from those resistors and - at the same time - maintain a pretty low tempco, for probably less than $10.

5. What about the vishay resistor networks ACASA1002U1002P1AT, ACASN1002U1002P1AT? 0.1%/10ppm abs., 0.05%/5ppm relative and really cheap compared to MPM etc.

6. Did anyone (with more analog knowledge than me) put some thought into the 1A range (e. g. in a similar fashion as the Keithley 238)? I'd find this device even much more useful with a 1A @15-30V range, even if it wouldn't be quite as precise as the lower ranges due to the significant power dissipation in the shunt.

Thanks,
Sebastian

[Kleinstein]

The 1 A range would likely need switching with a mechanical relay. The leakage than would not be a problem and one may also consider a realy for the 100 mA range.
Already the 100 mA range shunt gets quite warm and really should be large size, so more like a heat sink mountable power resistor like the PBH series.
8 film resistor in parallel may work, if there is plenty of spacing (e.g. 10 mm) in between, but this would add trace resistance and would thus need extra copper. 0.5 W is quite some heat for a precision shunt.
If using the 1 A range one would kind of need a lower burden voltage (e.g. 0.5 V range)  and lower noise amplification (especially avoid the 100 K resistors at the difference stage) for the 1 A range and likely use this also for the 100 mA range.

The ACASA resistor networks may work, though also not the best performance. There are multiple alternatives for the resistors, including ready made difference amplifiers at places.

[SebastianH]

The 1 A range would likely need switching with a mechanical relay. The leakage than would not be a problem and one may also consider a realy for the 100 mA range.

Exactly my thought, although for the 100mA range the Photo-MOS relay might be ok depending on the personal requirements In theory it would be best to use latching relays to avoid additional power dissipation. Could this lead to problems at start up, because effectively one will end up with an undefined state?

Already the 100 mA range shunt gets quite warm and really should be large size, so more like a heat sink mountable power resistor like the PBH series.
8 film resistor in parallel may work, if there is plenty of spacing (e.g. 10 mm) in between, but this would add trace resistance and would thus need extra copper. 0.5 W is quite some heat for a precision shunt.

Probably true, especially with heatsinking (tempco of those resistors isn't always all that great, partially as bad as 50ppm if I remember correctly). Then again, if I'm really going to build this (which I really want to try), I'd be very restricted regarding board height for the analog board (<2 cm), whereas length/width don't matter that much. Therefore I'd probably end up with SMD solution.

 

If using the 1 A range one would kind of need a lower burden voltage (e.g. 0.5 V range)  and lower noise amplification (especially avoid the 100 K resistors at the difference stage) for the 1 A range and likely use this also for the 100 mA range.

Yes, for the reasons discussed I'd propably end up using another array of 1206 10ppm 0,4W (something like 4 strings with 2 resistors each), 0,5 Ohm total. I think since the 1A range is a somehow just a bonus, it wouldn't be too bad if the performance was slightly worse. I'd likely also use a buffer and 10 k resistors for the amplifier. Maybe one could get away with using the Low Range for the FDAC (setpoint) and the x8 Gain of the AD7190  No idea how much "precision" would be left, but that way, no addtional hardware would be required.
However, the main difficulty (to me anyway) would be the lower voltage/high current amplifier design itself, I think.

The ACASA resistor networks may work, though also not the best performance. There are multiple alternatives for the resistors, including ready made difference amplifiers at places.

Yeah, there are certainly even better options, but value per money is there, I think. If I do my own layout, chances are high that this will be the main weakpoint (starting with simple prototyping mistakes, best practices, ...). (Although I'm an electrical engineer, my work has always been completely unrelated to electronics, metrology aso.; hence I consider myself a advanced beginner).

If you're refering to INA ICs: I haven't found one, that has comparably
- low input bias current
- low offset drift
- supply voltage ratings
- available in diy-friendly packages

all at once (and also, it has to be available in that package). But I'm open to suggestions. As a side note: I'm aware that I can't just compare all datasheet values of an INA with an opamp, because the circuit of 3 opamps will obviously different from one single opamp. However, very many INA have input bias currents in the 1nA range, which I find quite high for obvious reasons

Thanks,
Sebastian

[Kleinstein]

A latching relay should be ok. The SMU would start with the voltage /current set low from the DAC anyway. There should be no unsafe relay setting and the undefined state would only be for the first 100 ms or so. Autoranging with the current is anyway a bit tricky, as switching the shunt will cause transiensts for the control.

Extra gain for the higher current shunts would add a little extra hardware, like an extra OP-amp for some gain and a CMOS switch like DG419. So some extra effort, but not really much. On the other side it saves at the shunt resistors, that could get away with lower power rating (e.g. 1/4) or get better performance.

With the INAs / difference amplifiers the current chip availabiltiy makes this even less attractive as one would rely on a relatively special chip. 4 resistors and an OP is easier to get.

If hight is an issue, that multiple SMD resistors are definitely a good idea, but they still need area.

The current plan uses a 5 V full scale for the voltage and current signals at the cross over circuit part and scales the DAC and ADC signal accordingle with a factor of some 2.2 and 0.45. It may be easier to work with a slightly reduced voltage range, like 4 or 4.5 V and than have the scaling for the DACs and ADC input with a factor of 2 and 0.5, which is easier.
For the resistors there are also resistor arrays like NOMCT with 8 x 10 K  that could replace several pairs of resistors.
For the resistors there would be even the option to use just seprate good quality resistors. With parts from the same batch the matching is not that bad either - not guarantied, but usually the matching is quite good. With the arrays one often also only gets typical matching that is really good.

[SebastianH]

I'm currently working on the schematic which (for the most part) is based on both Dave's and Jaromir's design, except I'd really like to implement a 1A (@15V or so) range. Maybe you can give me some input on some parts of my current schematic.

Current Ranging:



Main changes as previously discussed in this thread:

• "Full" instrumentation amplifier + 10k/10k Vishay ACASA 10ppm resistor array (should be good enough for my first attempt)
• 1 mA range with dedicated switch for the sense line
• Array of multiple 1206 resistors, 10 ppm, 0.4W, for both 100mA and 1A range, mostly to limit board height. The shape of this array (like in 4x2 and so on) is not fixed yet. Thermally it would be optimal to leave quite some space between the resistors. For the 100 mA range with its 50 Ohm total shunt resistance the copper tempco of the traces should add no more than 5ppm or so even if I leave *some* space between them (I think), which would be acceptable. For the 1A range though, with its 0.5 Ohm shunt resistance things are way more complicated. To my knowledge I can't get cheap (or actually any...) 10ppm thin film resistors below ~2 Ohms (especially right now). This reduces possible series/parallel combinations of resistor networks drastically. Additionally, the total value of only 0.5 Ohms leads to many additional ppms in the copper traces, especially when leaving some more space between the resistors - and for the circuit shown with that many parallel resistors (10 for 4.99 or 11 for ~4.5 Ohms), I believe there is no great solution for the problem. I might end up bringing those resistors a little bit closer together as I orginially planned and rely a little bit more on the low tempco of the 10ppm resistors itself. My goal with this shunt would then be something like 25 ppm. I also thought about a low profile heatsink with a thermal pad, either on the bottom side of the pcb by using thermal vias or even on top of the SMD resistors with a thick enough thermal pad; I'd assume this wouldn't have any significant impact on the electrical properties of the fairly low resistance shunt (?). Anyways... I might or might not lower resistance values to ~4.53 *Ohm and use 2:1 and 1:2 ratio for ADC and ADC signal conditioning as Kleinstein recommended. Those are still available at mouser for the most part, believe it or not. I have some 4.99 *Ohm precision resistors left, so the advantage for me personally would likely be limited. And to the opposite, the current approach - while less "clean" - leaves more options to finetune the resistance values and potentially a bit more dynamic range. Not sure useful this is though.
• EE2/EC2 relay for 1A range.

1A range:

For the 1A range I followed the approach of the Keithley 238. Unfortunately, as far as I know no (full) schematics are available publicly beyond what is described in the service manual (which is easily accessible). Hence I have to apply my limited knowledge here. Note: If you haven't looked at the service manual of the 238 in the past, it might be helpful to understand what I'm talking about  , otherwise a short intro to the topic: The 238 has two amplifier stages: 100mA (0 .. +-110V) and 1A (0 .. +-15V). The 1A amplifier has its own +-30V/1A power supply and is driven by the 100mA amplifier. So the first stage seems to drive both the load (pretty much as it would in the 100 mA range, a small part of the load anyway) and the second stage. The difference in the 1A comes from the the second amplifier that "injects" its current into the 1A current shunt (which is connected in series with the 100mA shunt, as can also be seen in my schematic).

The service manual states: "For low output currents (0 to 20mA), the 1nA-100mA stage is used. As the current is increased, the transistors in the 1A output stage start to provide output current up to 1A."

Please correct me if I'm wrong, but together with Figure 4-3 of the service manual (yeah, it's only simplified) and the excerpt from the service manual I get the impression, that the second stage is indeed a class B amplifier that phases in slowly, probably due to the V_BE of the darlington transistors used, because otherwise I would expect the amplifier to immediately "support" the first stage and not only after 20 mA. My understanding is further, that those "20 mA" develop the voltage accross the 100 mA current shunt that leads to the voltage differential between (a) the ground of the +-170V and +-30V/1A (negative output terminal) and (b) the output of the first stage and that is necessary to overcome V_BE of the darlington transistors of the second output stage. Did I get something completely wrong or would Keithley actually use such a design? To be honest, I didn't really look to closely into the operation of the 236-238 except for the parts relevant for the 1A range, so who knows. I think class B would make the life a lot easier. That being said, I'm not sure I'd really want to have this distortion.

What I did was to pretty much just copy and past the first stage stripped some things away and added some sort of a biasing circuit.



The service manual shows a Zener-based biasing circuit for the Mosfets that references the floating +-15V power supply like in the next screenshot. This will "lift" the gate voltage above the +30V rail (or below -30V) whenever the output comes close to the rails. I'm not so sure that something like this would be very useful in my example.



I'm in the a completely new territory and I definitely encourage improvements and concrete ideas. Btw: I sometimes just used a fairly random part from the library, like the MOSFETs. But still feel free to make suggestions

-Sebastian

[Kleinstein]

2 separate power amplifers for the low current and high current is a possibility, but not really needed.  It should be possible to use a configurtaion somewhat similar to the newer Keithly 24xx. There is a crude and buggy schematics in the service manuals, that at least gives the genral idea:  similar to the current schemantics use a king of class G (sometimes also called class H) output amplifier that uses power from 2 power rails (e.g. +-150 V and +-15 V). This way the sink mode would also generate less heat as only the lower voltage is added.
Of cause a more complicated amplifier also has more chances for misbehaving (e.g. oscillate).

The power transistors (MOSFETs) with sufficient SOA to provide 100 mA from a high voltage are usually also OK to provide 1 A from some 15 V.  So there is no real need a fully separate output stage.

A separate power stage and switching with a realy is of cause possible and would get less voltage lost at the power transistors in the 1 A range. In this case I would prefer just turning off the unused amplifier and not have them in series, which looks odd.

For the resistors one may try a test upfront, how good the resistors actually perform. With the SMD resistors the way they are mounted can have quite some effect on both the thermals and also the TC / drift. It is also not just the TC but also possible thermal EMF from temperature differences at the ends.

The shunt switching looks OK.  I would consider to wire the next 1 or 2 current ranges (500 Ohms and 5 K) also in series for the sense part and this way save 2 more of the DG444 switches and avoid there leakage.

[SebastianH]

Thank you for pointing me to the Keithley 2400 service manual, I appreciate all this input.

To be honest, a class G/H approach was my first thought, since this seems to be a much more logical and cleaner way of doing this. The problem for me is the complexity you mentioned - I might need an additional (higher voltage) bias supply, potential instability issues and what not. And with the two-staged approach the 1A is just an add-on - if for whatever reason I can't get the second stage working or just want to start with as little complication as possible, I could leave that circuit unpopulated and still have a working 100mA SMU (ideally). I'll give this another thought though.

The 2400 service manual is very interesting: "The output stage drive transistors are biased in class B configuration to prevent the possibility of thermal runaway with high-current output values." If that's good enough for Keithley, so it is for me. Power consumption/quiescent current and thermal runaway was something I was already concerned about with "my" design

For the two-staged approach I find it a bit difficult to find the right solution to disable/switch off the second amplifier. Keithley used one n-channel and one p-channel MOSFET for the positive and negative output darlington pairs respectively on the input side. They biased the cascode MOSFETs with a zener supply referenced to the floating circuit. My understanding would be that this biasing also ensures that the aforementioned MOSFETs (used to isolate/disconnect the input) can be of a fairly low voltage rating. Otherwise a large voltage could develop between the output of the first stage (floating common - +-150V roughly) and the input stage of the second when operating in the low-current-high-voltage ranges. But I think this would also require to open (relay) the connection between the output ground of both the -+170V and +-30V output power supplies - and therefore to have the second amplifier and its +-30V power supply floating like the control circuit itself. Would you agree?

Thermal EMF is a good point, which I considered briefly. I had a look at Microchip AN1258, where they have an example of an 1206 resistor. They use what looks like examplary values that are easy to work with, but should be in the correct order of magnitude. Having a 10°C/inch gradient accross this resistor axially would result in a 1.2°C gradient accross the resistor and a thermal voltage of almost 40µV. This would yield an additional 8ppm error for a full scale value. In this case, the resistors are the heat source, so in a perfect world a single resistor on a pcb shouldn't have a huge thermal EMF issue, I would think. In an array it might be a different story depending on the actual component and trace layout. In theory I could compensate some of those effects by means of a careful layout, but I also have to keep in mind not to increase the trace resistance too much because of the copper tempco. And then there are external heat sources as well, and even some air flow potentially.

Not quite sure about this, but I believe this might be a good point for me to not over overcomplicate things - I know at some point there will be limitations to my approach. And if i remember correctly even Keithley degrades their performance specs for both the 100 mA and the 1A ranges in a very significant way 

I assume you meant something like this with the series connection of the higher range current shunts:



With this arragement, the 10 mA probably will become more susceptible to the thermal EMF problem. The question then is, which ranges are more important? And is the lower end of the µA range good enough to begin with, that 1/8th of the leakage currents from the switches makes the difference to compromise a bit more on the 10 mA range?

[wizard69]

Dave!

I must say I'm most impress with your build here.   I pretty much dropped out of the thread sometime ago due to only really needing a portable precision DC voltage source.   However today I spent a bit of time on your web site and have to offer up a big THANK YOU.   The thank you due to there being so much to learn on that site.   

Dave

[Kleinstein]

Switching off the separate 1 A amplfier could indeed be a bit tricky, as the the output is via the raw source (e.g. transformer), and not a simple amplifier with the output from the transistors. It can at least be confusing and hard to follow in the head. So definitely a part to simulate to be sure, not to get unexpected high or reversed voltage in some cases.

The tendency for the SMU is to have a relatively slow regulation compared to a more normal lab PSU. So there are compromises with precision and wide range versus speed.  The design calls for relatively large emitter resistors at the output stage. So even if biased like a class AB amplifier thermal run-away would be a lesser issue and the control loop will have to do some work to compensate anyway. So a true class B would not be much slower.

With only a small current flowing the 1 A and 100 mA range shunts sould not contribute much termal EMF. The 10 mA range already has the larger voltage drop for full scale and thus less sensitivity to thermal EMF anway. To keep the thermal EMF effect low a relatively symmetric design around the shunts that get hot helps. Having more shunts in parallel just averages to thermal EMF, so it more like makes things a bit easier.

Even in the series configuration for current range switching  there are still 6/5 CMOS switches and the PM relay to give some leakage. So 1 or 2 CMOS switch less does not make such a big difference, but it does not hurt.

With the measurement and control signals in the +-5 V range with maybe a little overrange I wonder if one really needs a +-15 V supply for the OPs. One may get away with a lower supply like +-10 V and thus less heat loss. This may be a point is compact, low power version.

[jaromir]

Array of multiple 1206 resistors, 10 ppm, 0.4W, for both 100mA and 1A range, mostly to limit board height.

That is a good approach. Let my add a bit of observation from my SMU:

The analog board I'm using is not the first version. The respin was due to multiple reasons, one of them being drift of current output, especially on 100mA range (much less on the 10mA range, negligible on the other ranges). That pointed at self heating temperature drift of the sensing resistors. To remedy this, I made a few changes, ordered by perceived importance:
1, Decreased voltage drop at sense resistors from original 5V to 2V (compensated by increased gain of the shunt sense differential amplifier)
2, Divided the current across four resistors for the 100mA range, as opposed to two resistors in the first version
3, Changed the PCB layout: resistors got copper planes to help with the heat dissipation (see attached picture)
4, Used resistors with 15ppm/C tempco, as opposed to previous 25ppm/C.

Those changes caused the thermal drift to decrease from ~60ppm to ~17ppm, in both cases loaded with 90% of maximal output.

[SebastianH]

Hey Jaromir,

great to hear from you as well, really impressed by your SMU!

1, Decreased voltage drop at sense resistors from original 5V to 2V (compensated by increased gain of the shunt sense differential amplifier)

At first I didn't want to use the 2*10^x Ohm values, partially because I didn't feel comfortable with the 250k required with the initial design, but since I decided to buffer both inputs of the diff amp this is not an issue any more.
I think I'll even combine this with Kleinsteins approach, given that appropriate 10 ppm resistors are actually available. I could use ~2.2*10^x Ohm shunt resistors and a gain of 2 for the instrument amplifier and use a gain of 2 and 1/2 for the ADC/DAC signal conditioning. And maybe the having the capability for some 5-10% overrange is more a feature than lost dynamic range  I still use multiple resistors.

So even if biased like a class AB amplifier thermal run-away would be a lesser issue and the control loop will have to do some work to compensate anyway. So a true class B would not be much slower.

What would you consider a high emitter resistance in case of the 1A range? I'd like to use something like 0.5 Ohms to limit power dissipation.

With only a small current flowing the 1 A and 100 mA range shunts sould not contribute much termal EMF. The 10 mA range already has the larger voltage drop for full scale and thus less sensitivity to thermal EMF anway. To keep the thermal EMF effect low a relatively symmetric design around the shunts that get hot helps. Having more shunts in parallel just averages to thermal EMF, so it more like makes things a bit easier.

Not sure I can follow. The scenario would be that I've used the 100mA/1A range and kind of heat-soaked the unit a bit. Then I switch to the 10 mA range and potentially see a series (thermal EMF) voltage of say 50µV from those series shunts. I'm certainly not sure, how large the thermal EMF voltage would be, but still. Without this "heat-soaking-process" it wouldn't matter that much, I'd agree - or am I missing a fundamental concept here?

Even in the series configuration for current range switching  there are still 6/5 CMOS switches and the PM relay to give some leakage. So 1 or 2 CMOS switch less does not make such a big difference, but it does not hurt.

Agreed.

With the measurement and control signals in the +-5 V range with maybe a little overrange I wonder if one really needs a +-15 V supply for the OPs. One may get away with a lower supply like +-10 V and thus less heat loss. This may be a point is compact, low power version.

I thought about that, but for other reasons. Right now I'm not sure how to design the power supply - I guess this will be something for a later development stage for me. But due to space restrictions I likely have to either wind my own transformer (/order a custom one) or use a high quality DC/DC converter module like Dave did (certainly not a separate off-the-shelf line transformer for the floating +-15V bias circuit). In case I end up with a DC/DC converter I was wondering whether to use a linear regulator after that. At the usual operating frequency of those isolating DC/DC converters - that will be as high as 300kHz (not to mention all the higher frequency switching noise) - the linear regulators don't do much, but still a little bit, maybe ~10dB PSRR. Is that still better than nothing? (I would think a lower operating voltage of the opamps might reduce the PSRR, so I don't know). This would give me something like 10V-12V. And certainly, the power dissipated in the linear regulators could be sinked into the case or even better the heat sink.
Anyway, I like this idea. And I'd love to find a company for custom transformers at a reasonable price.

[Kleinstein]

DCDC converters may introduce quite some common mode interference. So I would not consider this very attractive - at least not most of the of the shelf ones.

Thelinear regulator will not help very much with the regulation at the higher frequencies. However it keeps the voltage stable and allows more series resistance in the filter.
How good the LDO it also depends a lot on the output capacitor and the layout around the capacitor.
Unless the supply gets really low, there should be relatively little difference in the PSRR between 2x15 V or 2x10 V supply. It is more that less heat in the OPs that can reduce thermal effects.
One may still have to check the maximum signal level needed.

The capacitive dropper idea for the auxiliary +-15 V at the power stage is a good idea for low power. No sure if already mentioned: the capacitive droper could use full wave rectification and thus get away with smaller (e.g. half the size) capacitors. For the extra 1 A stage this would not longer be practical.

For the emitter resistors some 0.5 ohms are about the minium for the TIP41/42. Those resistors are not only there for the hard wired fast current limit, but also effect the transfer function. More restance may make it more stable, while drift of these resistors is not critical. The smaller the resistor, the better the thermal coupling in the amplifier / current stabilization should be. Compared to the more normal audio amplifier the TIP41/42 see relatively little voltage and heat as most of the heat goes to the MOSFETs. So even with only 0.5 ohms the situation should be a bit more relaxed than with many audio amplfiers.

[SebastianH]

DCDC converters may introduce quite some common mode interference. So I would not consider this very attractive - at least not most of the of the shelf ones.

I wouldn't even consider those if custom transformers were readily available, adds so much complication to achieve the same performance. Dave himself did quite an elaborate test on those and found a good correlation between cost and common mode noise and so on. There are also some "medical grade" DC/DC converters (i. e. specified for MOPP standards and so on) that are optimized for low patient/ground leakage (supposedly in the µA range), was thinking about testing the performance of one of those.

Thelinear regulator will not help very much with the regulation at the higher frequencies. However it keeps the voltage stable and allows more series resistance in the filter.
How good the LDO it also depends a lot on the output capacitor and the layout around the capacitor.
Unless the supply gets really low, there should be relatively little difference in the PSRR between 2x15 V or 2x10 V supply. It is more that less heat in the OPs that can reduce thermal effects.
One may still have to check the maximum signal level needed.

The capacitive dropper idea for the auxiliary +-15 V at the power stage is a good idea for low power. No sure if already mentioned: the capacitive droper could use full wave rectification and thus get away with smaller (e.g. half the size) capacitors. For the extra 1 A stage this would not longer be practical.

I included this in my schematic a few days ago, but I don't believe that it was discussed previously. The nice thing is, it's just a matter of connecting the diodes differently. I think it is worth mentioning that with shunt regulation there will be some ripple on the rails no matter what. With a 15kOhm load/1mA and a random 1W Zener diode from the library (KDZ15B) for example in the range of about 50mVpp (1µF-20% for the "AC capacitor", 220µF bulk capacitance). This could be reduced to about 25mV with only 100µF (however at a higher frequency -> PSRR). I'd be really interested to hear, what Jaromir/Dave are measuring in real life. I might use a 680n-1µF AC capacitor, a higher voltage clamp (24V or so), 47µF bulk capacitors and some SMD linear regulators. Might be a bit overkill, but why not... (one reason could actually be the quiescent current of the linear regulators, which loads down the capacitive power supply a bit, not a real problem with the full bridge).

Btw: Hadn't looked at a shunt regulator after a rectifier before that, so I can recommend this to anyone that is in the same shoes - the waveforms are pretty self-explanatory, but interesting still...

For the emitter resistors some 0.5 ohms are about the minium for the TIP41/42. Those resistors are not only there for the hard wired fast current limit, but also effect the transfer function. More restance may make it more stable, while drift of these resistors is not critical. The smaller the resistor, the better the thermal coupling in the amplifier / current stabilization should be. Compared to the more normal audio amplifier the TIP41/42 see relatively little voltage and heat as most of the heat goes to the MOSFETs. So even with only 0.5 ohms the situation should be a bit more relaxed than with many audio amplfiers.

Ok, I'd probably use Daves/Jaromirs design for now, for the TIP41/42 I think they used more like 22-27 Ohms. The 0.5 Ohms would be for the darlington pair (1A range). Yes, that influences the output/input impedance of the amplifier as well.

[RikV]

Dave, would you consider to implement an Ohms measurement function? Both Four wire constant current (basically FIMV with calculated output value) and constant voltage (FVMI)? This instrument is capable of precisely measuring very low and very high resistance values. I know thes caculations can be made over SCPI but real-rime readout is also very interesting.

SebastianH: When can we expect to see some of your schematics? I am very curious!

[SebastianH]

Maybe a first version at the end of the week? But I have to limit your expectations... It will be very close to both Dave's and Jaromir's schematics, with some input from mainly Kleinstein, Roehrenonkel and myself.

I'm still not sure whether to use a buffer for the positive reference. A buffer may help, but it might not be necessary or even decrease performance if not implemented correctly. I'd keep the capacitors at the REF input pins in any case (probably 10µF in series with a low value resistor to ground + 100nF directly accross VREF+ and -), for the VREF sampling which I believe happens at the modulator frequency exactly as the input sampling. I might just add a footprint and a solder jumper for some tests.

There is a PGA in the AD7190 (x8, ...), at the same time, I have 2 ADC inputs left. I could use one of those for a x10 current measurement specially for the 1A range. Would an external solution give me much better performance? Then I needed another switch, because I wouldn't feel comfortable with just driving the output to the positive rail all the time. Especially for zero offset/zero drift/chopper opamps like the OPAx388 with their parasitic diodes accross the inputs, I think it's a pretty good idea to keep them in the linear ("v+ - v- = 0V") region. Then again, could I get away with it due to the pretty high feedback resistor (for x10)?! Anyway, I personally don't like it What do you think?

[Kleinstein]

The DC accuracy is anyway limited by the use of multiple non AZ OPs in the signal chain. This effects the DAC, the ADC and the amplifiers for current and voltage sense. So the PGA gain at the ADC is of somewhat limited use. One might consider lower noise/drift amplifiers at the ADC inputs.
One of the inputs for the currents sense needs to be very high impedance for the low current to work. So this kind of has to be the OPA140 or similar.

I thinks for the amps part one could add an extra switch (e.g. DG419 or 2/4 of DG444)  in the feedback to switch the amplifier for current sense between the gain 1 for the lower currents and maybe a gain of some 5 when using the higher currents like 100 mA and 1 A.  The OP-amp for the other side could be a different type (e.g. OPA207 or OP27) with less drift. The extra gain would than not only help the read out but also the regulation.

Dave, would you consider to implement an Ohms measurement function? Both Four wire constant current (basically FIMV with calculated output value) and constant voltage (FVMI)? This instrument is capable of precisely measuring very low and very high resistance values. I know thes caculations can be made over SCPI but real-rime readout is also very interesting.

SebastianH: When can we expect to see some of your schematics? I am very curious!

The performance for low ohms may be limited, as the voltage read-out is not especially accurate / DC stable. A reasonable high resolution DMM may have something like 10 x better votlage readings for low ohms and thus could get away with 1/10 the test current. To get around this at least a little it would be more like low frequency AC for the low Ohms testing.

[SebastianH]

The DC accuracy is anyway limited by the use of multiple non AZ OPs in the signal chain. This effects the DAC, the ADC and the amplifiers for current and voltage sense. So the PGA gain at the ADC is of somewhat limited use. One might consider lower noise/drift amplifiers at the ADC inputs.
One of the inputs for the currents sense needs to be very high impedance for the low current to work. So this kind of has to be the OPA140 or similar.

I thinks for the amps part one could add an extra switch (e.g. DG419 or 2/4 of DG444)  in the feedback to switch the amplifier for current sense between the gain 1 for the lower currents and maybe a gain of some 5 when using the higher currents like 100 mA and 1 A.  The OP-amp for the other side could be a different type (e.g. OPA207 or OP27) with less drift. The extra gain would than not only help the read out but also the regulation.

Let me confirm what you have in mind. Replace say the Opamp buffering the positive input of the difference amplifier by an OP27 (which has slightly less offset, drift and noise than the OP145). Then add two switches to the differential amplifier (either 2x DG419 or two switches of a total of 4 in the DG444 package) to make the gain switchable? I didn't want to switch the gain on a differential amplifier because of the complexity of two switches instead of one for a simple inverting amplifier - and potentially differences between those two switches. Do you think this is unsubstantiated?

My plan was to add a DG444 for the integrator to have different gain settings for the different ranges. At least in my simulation I haven't found a "one size fits all" approach, at least none which showed a fast response for all ranges.

[Kleinstein]

The idea is to change the OP used as the extra buffer (not in daves plan, U304A in the schematics from june 12) from just a buffer to alternatively also work as amplifier relative to the other buffer. The switching would be a single DG419 for the inverting input of the OP. So there is essentially no added error or drift from the switch. The amplifier is than a little more like an INA with the optional gain already in the first stage, though only on one side.

Changing from an OPA145 to OP27 would be a seprate point to reduce the dirft / noise a little. Having alternative OP amps could be a good idea in several places. This could be for lower costs, better availabilty and in some areas better performance. This changes would be something to consider later, once the general circuit is ready.
One OP that may contribute quite a bit to drift and low frequency noise is the "amplifier" in front of the ADC (TLC2272 in Daves plan). I would consider a MCP6V77 in this place.

Switching the loop compensation (e.g. at the integrator) for different current ranges may be required or at least help. By nature the very low current ranges will be somewhat slow. Jaromirs plan uses switching of the emiter resistors at the output stage to change the gain (tranconductance) at that place. The switrching is no just to adapt the extra fast current limit, but also effects the gain and this way the loop compensation in the right way.
Another possible point may be a switch (e.g. photomos) for additional output capacitance (likely with some series resistance) for the higher current ranges.
As far as I see it the SMU will not have a very fast response time as wanted from a good lab supply. Chances are one would need to live with some compromises to allow the very large current range.

[SebastianH]

Your explanation did help somewhat, however I'm not sure why I would only use "one side" and how this should look exactly.
So, here is a INA, in this example Gain x2 from the second section and Gain 10x from the first section. In this case the switch in series with the gain resistor would introduce gain error. It might become irrelevant at low gain and large R9 and R10 resistors, and hence a large gain resistor R11 as well; or with low R_on switches/relays.



Do you mean by "one side" to just make U7 a simple non-inverting amplifier with switchable gain? As far as I can tell I would need two switches to avoid any gain error: To switch either the output directly (switch 1) or to switch the divided output signal to the inverting input of U7 (switch 2)?
Or did you mean to essentially make R10 a short, i. e make U7 a non-inverting amplifier, but reference the voltage divider to the output of the buffer U8?! Also Maybe I'm the only one to not understand this right away? Sorry

What makes you want to choose the MCP6V77 over say the OPA388? The MCP6V77 seems not to be available in a SOIC8 package, which I much prefer (try to stay away from those tiny packages as much as possible).

Good point regarding the current limiting circuit, have to look into that.

Sebastian

[Kleinstein]

R10 as a short is exactly my idea. So that side still works as a buffer.  Switching the gain is between 1 and one value is than only needs a SPDT switch.
To get a gain of 1 one would need the 2nd stage as the normal gain of 1.

The MCP6V77 is compratively cheap - I had not considered the case to be that critical. The OPA2388 (as a dual) is currently a bit hard to get. Other AZ OPs may be possible too - that amplifier does not have to be super low noise. One could even consider to skip these amplifiers alltogther and use a simple 1:1 divider towards a buffered 2.5 ref. level instead. That solution would even be a bit less sensitive to the resistor quality. It should however use the ADC internal buffers and add some INL this way.

[SebastianH]

Ok, I assumed the DG419 is SPST, and it's always dangerous to assume something Now everything makes much more sense.
Would be nice to use the spare DG444 SPST switch, but I'd need two of those to replace the DG419. I could potentially even connect the 1mA shunt in series. Hmm. Opinions?



Yes, unfortunately, I would have to use the single channel version OPA388 at the moment. Not too happy about that.

[Kleinstein]

The 1 mA range should still be OK with the series circuit: the noise of some 2.2 K or 4 K is still not too much compared to the OP-amps. The question is more if the 100 µA range needs a separate sense or maybe a lower resistance switch. This would especially the case with one 2.2 K and reduced voltage also for the lower currents.